The mouse has a Polycomb-Wke chromobox gene

During the early stages of regulative development, embryos assign positional values to their parts. Wolpert has suggested, in his influential theory of positional information (Wolpert, 1969), that this specification is achieved by a morphogen concentration gradient. Mathematical studies have concluded that such gradients can only lay down a pattern over a distance of about one millimetre, or about 100 cell widths (Slack, 1991). This estimate corresponds very well with the actual size of embryos undergoing positional specification: the anteroposterior (A/P) axis of the Drosophila embryo at cellular blastoderm stage, the mouse embryo at primitive streak stage, and the Xenopus embryo at gastrulation each cover some 100 cells (Rugh, 1968; Nelsen, 1953).

The morphogen needs a cellular mechanism that can convert its concentration into fate. The Antennapedia (Antp)-iike homeobox gene family can fulfil the requirements of this mechanism. In situ hybridization on mice and Drosophila embryos has shown that, at the time of positional specification, the members of this family form a series of different but overlapping domains of expression along the A/P axis (Gaunt, 1991). Each area of the embryo therefore contains a unique blend of Antp-like homeobox proteins. Experimentally altering the blend causes the area to adopt an abnormal fate (Kuziora and McGinnis, 1988). This suggests that normal cellular fates are specified by normal levels of the An/p-like homeobox proteins.

With a morphogen and a set of Antp-like homeobox genes, a small embryo can assign positional values and fates to its parts. However, it faces a problem: it needs to grow. Due to their physical properties, the concentration gradients that provide the initial positional information cannot continue to supply this to the expanding embryo. It is therefore necessary for the parts of the growing embryo to commit their newly specified positional values to memory. A consequence of this operation is that, when part of the embryo is moved to an ectopic site, it differentiates as it would have done if left undisturbed. Classically a piece of an embryo that behaves in this manner is called determined, and the process of committing positional value to memory is known as determination.

Determination has been shown to occur in many organisms. In chickens, when segmented or unsegmented somitic mesoderm is moved from cervical regions to thoracic regions, it differentiates to form vertebrae of cervical form (Kieny et al., 1972). Transplantation experiments on amphibian neurulas have revealed patches of determined mesoderm, the so called secondary fields (De Robertis et al., 1991). These patches form such features as the ears, the balancers, the gills, the heart and the limbs. Insects also undergo determination. When Drosophila blastoderm stage cells or imaginal discs are transplanted, they differentiate in a manner appropriate for their donor sites and not their host sites (Simcox and Sang, 1983; Gehring, 1967).

A group of Drosophila mutants, the Polycomb (Pc) group, has problems in fixing and maintaining the determined state (Jiirgens, 1985). These mutants can specify their initial positional values but are unable to remember them (Dura and Ingham, 1988; Struhl, 1981), a forgetfulness that leads to larvae and flies made up of inappropriate parts. Strong mutants of the groups namesake, Pc, have initially correct patterns of Antp- like homeobox gene expression, but at the extendedgerm band stage (stage 11, ∼6 hours) these become indiscriminate (Kuziora and McGinnis, 1988). This results in all the segments of the larva adopting the fate of what appears to be the systems ground state, the abdominal 8th segment (Wedeen et al., 1986). The switch from correct to aberrant patterns of Anlp-like homeobox gene expression in Pc mutants suggests that Drosophila, having established its positional values, commits these to memory at the extended-germ band stage by tying down the states of activity of its Antp-like homeobox genes.

A clue as to how the Drosophila memory mechanism may operate has emerged from the cloning and sequencing of the Pc gene (Paro and Hogness, 1991). The Pc protein shares a short region of homology with Drosophila Heterochromatin Protein 1 (HP1), a component of heterochromatin that plays a role in the position effect variegation (PEV) phenomenon (James and Elgin, 1986). This finding has led to the proposal that the Antp-like homeobox gene expression patterns are fixed by preserving the active genes in an open and competent chromatin state, whilst encapsulating the inactive genes within heterochromatin-like complexes, thereby rendering them inexpressible (Gaunt and Singh, 1990; Paro, 1990). Clonal inheritance of these different chromatin states maintains the expression patterns of the Anip-like homeobox genes through time, so maintaining the determined state.

The region of homology between the Pc protein and HP1 shares 65% identity over 37 amino acids and is called the chromodomain (c/iromatin organization modifier). Zooblot analysis has shown it to be conserved across the animal and plant kingdoms, and a number of genes from the mouse and man have been cloned and shown to contain chromodomains (Singh et al., 1991). The probe used to isolate these initial clones was derived from an HP1 clone. It is perhaps, therefore, not surprising that the genes characterized so far appear to be more similar to HP1 than Pc. To see if genes of a more Pc type also occur in the mouse, a finding that might suggest a broadening of the applicability of the Drosophila memory models to other species, we have conducted a search using a probe derived from a Pc clone. It is the results of this screen that are presented here.

A 5 ′ primer (5 ′-GAA-TTC-TAC-GCG-GCT-GAG-AAA-ATC-3 ′) and a 3 ′ primer (5 ′-GGA-TCC-GTT-TAC-CTC-CGG-TTC-CCA-3 ′), which demarcate the Pc chromobox, were used to generate a PCR probe from a Pc cDNA clone (Paro and Hogness, 1991). The probe was used in a low-stringency screen of an 8.5-day mouse embryo library (Fahmer et al., 1987). Nylon filters (NEN; NEF-978) were hybridized overnight at 58°C in NEN’s alternative mix [1 M NaCl, 50 mM Tris-HCl (pH 7.5), 1% SDS, 10% PEG-8000, 5 × Denhardt’s Solution, 0.1% sodium pyrophosphate], with denatured salmon sperm DNA at a final concentration of 10 μg ml ⁻¹, and a probe concentration of 2 ×10^s cts minute ⁻¹ ml ⁻¹. After being washed twice for 20 minutes at 50°C in 2x SSC and 1% SDS, they were autoradiographed with intensifying screens for 10 days at −70°C. Positives were isolated as pure clones and their inserts subcloned into Bluescript KS+ (Stratagene). The inserts were sequenced, using the doublestranded dideoxy method (Sanger et al., 1977) described for Sequenase Version 2.0 (USB), by a combination of primer walks and directed exonuclease III deletions (Henikoff, 1984).

Because the M33 cDNA lacks an initiation codon, a genomic clone was isolated. An ES cell genomic library (a kind gift from A. J. H. Smith) was screened with a radioactively labelled 280 base-pairs (bp) EcoRI-Ps/I probe from the 5 ′ end of the M33 cDNA. The library filters (NEN; NEF-978) were hybridized overnight at 65°C in the NEN alternative mix, with a final denatured salmon sperm DNA concentration of 100 μg ml ⁻¹, and a probe concentration of 1 ×10^s cts minute ⁻¹ ml. They were washed twice for 30 minutes at 65°C in lx SSC and 1% SDS. After autoradiography at −70°C with intensifying screens, positives were picked and purified. The inserts were subcloned into Bluescript KS+ (Stratagene) ready for sequencing. A primer homologous to the 5’ end of the cDNA was used to extend the M33 sequence.

RNA was prepared by a guanidinium thiocyanate method (Chomczynski and Sacchi, 1987) and dissolved in a 40 unit ml ⁻¹ RNasin (Pharmacia) solution. Total RNA (15 μg) was separated by electrophoresis through a 1% agarose/2.2 M formaldehyde denaturing gel in MOPS buffer and transferred to Gene Screen Plus (NEN; NEF-976) by a capillary method (Lehrach et al., 1977). The filters were heated at 80°C for 2 hours then UV cross-linked (Church and Gilbert, 1984). The filters were prehybridized overnight at 60°C in the NEN alternative mix with a final denatured salmon sperm DNA concentration of 100 μg ml ⁻¹ and then hybridized, with radioactively labelled probes, in the same mix overnight at 60°C. After washing twice for 30 minutes at 60°C in 2 × SSC and 1% SDS, the filters were autoradiographed at -70°C with intensifying screens. The 280 bp M33 EcoRl-PstI fragment was used at 4 ×10^s cts minute ⁻¹ ml ⁻¹, and the filter was autoradiographed for 3 days. The β-actin control was used at 1 ×10^s cts minute ⁻¹ ml ⁻¹, and the filter was autoradiographed overnight. The cyclin B probe is a 156 bp Sau3a fragment from a genomic clone of murine cyclin B (Hamers and Singh, unpublished result). This fragment, which codes for an amino acid sequence that has been conserved between the cyclins of many species (MQNSCVPKK VLQLVGVMAM FIAS-KYEEMY PPETGDFAFV TNNTYTKH), was used at 3 ×10^s cts minute ⁻¹ ml ⁻¹, and the filter was autoradiographed for 3 days.

A ³⁵S-labelled antisense probe was generated from the 280 bp M33 EcoRI-Psil fragment using a previously described method (Gaunt, 1987). Methods for embryo sectioning, alkaline hydrolysis of labelled probes, in situ hybridization and autoradiography were performed as described elsewhere (Gaunt, 1987).

The 8.5-day mouse embryo cDNA library yielded 12 positives from 3 ×10⁵ plaques screened with the PCR generated Pc chromobox probe. Sequencing one of these, M33, revealed a 1.5 kb open reading frame. This clone does not contain an initiation codon, so its genomic locus was cloned, from an ES cell genomic library, and sequenced. An initiation codon was found immediately upstream of the cDNA-derived sequence. Some 50 bp upstream of this putative start site, there is an in-frame stop codon, suggesting that this is indeed the M33 protein initiation codon. The full-length M33 sequence encodes a 519 amino acid (aa) protein (Fig. 1). A 44 aa stretch of the N-terminal region of this protein shares 61% identity, and a further 16% similarity, with the chromodomain containing N-terminal region of the Pc protein (Paro and Hogness, 1991).

On aligning the M33 protein sequence with the previously described chromodomains (Fig. 2) two subgroups become apparent. The HP1 class is characterized by a block of negatively charged glutamic acid residues immediately upstream of the chromodomain and an amino acid sequence of the form LDCpdLI immediately downstream. Together, these features extend the region of homology between members of the HP1 class from 37 to 50 aa. In comparison, the members of the Pc class do not have the upstream glutamic acid stretch, and they possess a downstream identity, ILDpRLi, different to the HP1 class. HP1, M31 and M32 are members of the HP1 class; Pc and M33 are members of the Pc class. This classification extends to the sizes of the members of the HP1 and Pc classes: the HP1 class are all around 190 aa long, while both the Pc protein and M33 are considerably bigger, 390 and 519 aa, respectively.

Outside the chromodomain, comparisons between the predicted amino acid sequences of the chromobox genes (Fig. 3) have revealed a region of homology at the C terminus which is conserved within, but not between, classes. So, although M33 does not contain the Pc protein’s very noticeable polyhistidine blocks, it does share, in addition to the chromodomain, a block of 30 aa at its C terminus which has 53% identity and 20% similarity with the C terminus of the Pc protein (Fig. 4). At the whole protein level, M33 ′s high proportion of charged residues [basic (H, K, R) 17%, acidic (D, E) 9%] is comparable to the Pc protein’s charged residues [basic 20%, acidic 15%]. Immunohistochemisty has shown that the Pc protein is a nuclear protein (Zink and Paro, 1989) and three putative nuclear localization signals (NLSs) have been reported in its sequence (Paro and Hogness, 1991). Three potential NLSs also emerge in the M33 sequence when it is compared with the general features of an NLS, as defined by studies of the SV40 T antigen minimal NLS (Fig. 1) (Garcia-Bustos et al., 1991). M33 is therefore probably a nuclear protein, and a member of the Pc class of the chromodomain family. This assignment is strengthened by the M33 and Pc protein C-terminal homology.

The spatial and temporal expression patterns of M33 have been investigated using a combination of in situ hybridizations and northern blots. Embryos of 7.25, 8.25 and 12.5 days were sectioned and examined by in situ hybridization. Total RNA from 10.5-, 12.5-, 14.5-, 16.5-, and 18.5-day embryos, and newborn mice was separated by electrophoresis and examined by northern blot analysis.

In situ hybridizations to 7.25-day embryos sectioned within their deciduae showed labelling in both maternal and embryonic tissues. Maternal tissue labelling was confined to the outer layer of the decidua (Fig. 5A). Labelling of embryonic tissue was abundant within parts destined to form the foetus (ectoderm and mesoderm germ layers) but was not detected above background in the amnion, chorion or ectoplacental cone (Fig. 5B). Embryos of 8.25 days (Fig. 5C) maintained the distinction between labelled embryonic tissue and apparently unlabelled extraembryonic regions. At this stage, there was little or no specific labelling in the allantois. Embryos of 12.5 days showed labelling in all tissues (Fig. 6). None of the developmental stages examined showed anterior-to-posterior differences in the abundance of transcripts.

A northern blot of RNA samples from the final 10 days of murine embryonic development has been used to investigate the temporal nature of M33 expression. The blot clearly shows (Fig. 7A) that the level of expression gradually falls away from a high point at 10.5 days. One interpretation of this result would be that M33 expression is under developmental stage control. However, since M33 is a potential chromatin protein, the change in its expression may instead be due to a change in the rate of cellular division during this embryonic phase. This hypothesis has been tested by comparing the pattern of M33 expression with that of the murine cyclin B gene, a conserved ceil-cycle-specific gene whose expression level reflects the degree of cellular proliferation (Lehner and O’Farrell, 1990). The pattern of cyclin B expression (Fig. 7B) closely mirrors that of M33, both decline from a 10.5-day peak. It therefore appears that M33 expression is influenced by the embryo’s rate of cellular division, and not by its developmental stage. M33 has been seen to be expressed, at levels lower than in embryos, in adult tissues (data not shown), a finding in accordance with these tissues lower mitotic indexes (Altman and Dittmer, 1972).

The sequencing of M33 has led to the classification of two subgroups of chromodomain, the HP1 and Pc classes, and to the assignment of M33 to the Pc class. The intraclass homologies discovered in the C-terminal regions of the chromodomain proteins support this subgrouping, and may point to domains of functional significance. Although it has yet to be shown that M33 is a functional murine homolog of Pc, the data presented are not inconsistent with the hypothesis that M33 plays a role similar to Pc. The in situ hybridization study has revealed that M33, like Pc, shows ungraded expression. The Pc expression pattern is thought to be a reflection of the simplicity of its promoter (Paro and Hogness, 1991). Similarly, the embryonic M33 expression pattern may reflect its association with a CpG island, a simple vertebrate promoter.

Pc is a repressor of the Antp-like homeobox genes, and it maintains its repression through successive cell generations (Struhl, 1981). The molecular basis of this repression is unknown. However, two memory models propose that the Pc protein may participate in the formation of a heterochromatin-like complex that cloaks inactive Antp-like homeobox genes, rendering them inexpressible (Gaunt and Singh, 1990; Paro, 1990) . The Pc protein complex may, like the hetero-chromatin complex of PEV (Henikoff, 1990; Tartof et al., 1984), be able to spread along the chromosome. The models cast Pc in a simple role: it is a ubiquitously expressed building block. The model’s smart players are other genes that initiate spread of the hetero-chromatin-like complex at specific initiator sites.

In the context of the memory models, the conserved motifs at both ends of the M33 and Pc proteins may, we suggest, allow these proteins to act like interlocking building blocks, each addition spreading the complex further down the chromosome. The finding that the level of M33 is dependent upon the degree of cellular proliferation suggests that its gene product is required during the process of cellular division. This fits well with the proposal that M33 is a Pc-like gene since, according to the models, an Amp-like homeobox gene’s chromatin state remains static between divisions, but is clonally inherited at division in a process that requires a supply of the Pc protein and other heterochromatin-like building blocks.

There are hints in the literature that the chromodomain is not the only link between PEV and the memory mechanism. The memory models suggest that barriers to the spreading of a heterochromatin-like complex must form within the Amp-like homeobox gene clusters. The existence of such barriers would appear to be confirmed by the finding that the homeobox genes of the bithorax complex (BX-C) are resistant to PEV repression (Henikoff, 1990).

The memory models predict that an Antp-like homeobox gene control element removed from its chromosomal context will be able to set up an initially correct pattern of expression, but, if the disruption separates the control element from its initiator site, this pattern will not be correctly maintained. Dissection of the regulatory regions of the Ultrabithorax (Ubx) homeobox gene, a member of the Drosophila BX-C, has provided evidence which supports this prediction. The Ubx gene product is spatially restricted during embryogenesis to parasegments 5 to 13. Restriction of the Ubx protein to parasegment 5 is the role of the abx/bx domain. The abx subdomain has been incorporated into promoter-LacZ constructs to study its control elements (Simon et al., 1990). On its own, this subdomain is able to initiate LacZ expression with an anterior boundary coincident with the anterior edge of parasegment 5. However, it is unable to maintain this pattern. LacZ expression is seen to creep forward of parasegment 5 from about 9 hours onwards, until by larval stages all the imaginal discs are expressing LacZ. The inability of abx to maintain its expression pattern therefore suggests that abx may lack an initiator site. A similar defect in the memory mechanism might, we suggest, be the basis of the Antennapedia transformation. This homeotic transformation is due to ectopic head expression of the Amp gene from its P2 promoter (Jorgensen and Garber, 1987). Most of the mutants that give this phenotype have chromosomal inversions that map between Pl and P2 (Scott et al., 1983). In terms of the memory model, these inversions may separate P2 from its initiator site and thereby disrupt the correct maintenance of the P2 expression pattern, resulting in ectopic P2-driven expression.

According to the models, a cell’s positional address is preserved in the static chromatin state of its Antp-like homeobox genes. Having committed its address to memory in this way, the models predict that the state of activity of the Antp-Yike homeobox genes is cell autonomous. Consistent with this prediction, experiments with chick wing buds, where material from anterior regions was transplanted to posterior regions and vice versa, have shown that expression of the XlHbox 1 antigen is appropriate for the donor site and not the host site (Oliver et al., 1990). Interestingly, similar experiments looking at the expression of two non-Anip-like homeobox genes, Hox-7.1 and Hox-8.1, neither of which is a member of the Hox clusters, have shown that post-transplantation material adopts the host site’s expression pattern (Davidson et al., 1991). The finding that a Hox gene from within the Hox clusters preserves its state of activity on transplantation, but that Hox genes from outside the clusters do not, supports the notion that clustering and memory are associated mechanisms.

Errors in the transmission of the chromatin states at cellular division have been proposed to account for shifts in the fate of Drosophila imaginal disc cells (Gaunt and Singh, 1990). Shifts in fate can have dire consequences, as a cell in a new state may be unable to respond to its surroundings. A lack of responsiveness is thought to be a primary cause of cells becoming cancerous. It is intriguing therefore that a recently characterized member of the Pc-group, Posterior Sex Combs (Psc), has been found to have homology with the murine oncogene bmi-1 (Brunk et al., 1991; van Lohuizen et al., 1991).

The characterization of a Pc-like gene in mice extends the credible range of the memory models to include vertebrates. As the members of the Pc-group are characterized, it will be interesting to see if they have vertebrate counterparts. Support for the models, in either Drosophila or vertebrates, could come from an analysis of the chromatin state of the homeobox genes in different embryonic regions. The M33 protein product will be examined to clarify the degree of functional homology between it and the Pc protein, and a targeted mutagenesis of M33 is planned as it would clearly be of use in defining function.

The M33 sequence described in this paper has been submitted to the EMBL Data Library, and has been assigned the accession number X62537.

We thank R. D. Burton, Dr A. Ferguson-Smith and Dr W. Reik for technical assistance and Dr R. Paro for the kind gift of the Pc clone. J. J. Pearce is a recipient of an AFRC Ph.D. studentship and P. B. Singh is a Babraham Research Fellow.